General comment

The paper treats of ozone formation trend (2016-2020) due to shipping emission in China by using modelling simulations suggesting the relevance of this source on this pollutant. The topic is interesting and suitable for the Journal. However some aspects related to the choice done in modelling and to the interpretation of results are not completely clear or well described, see my specific comments. For this reason, I suggest considering the paper for publication after a revision step.

Specific comments

Anthropogenic emissions from other countries within the modeling domain (Table S2) was taken at 2010. It is possible to have a relevant uncertainty from this considering the period span of the study (2016-2020)?

Page 3, lines 1-4. It should be mentioned that there are also effects of titration of ozone due to ship emissions especially at local scale, a few kilometres, that could complicate both simulation and data interpretation see Merico et al (Atmospheric Environment 139, 2016, 1-10).

Page 3, line 6. Is this a sufficient resolution to investigate local processes leading to ozone formation? Generally, modelling of these processes is done using a much more refined scale.

Page 3, lines 31-32. What is Nm, nautical miles? Better to write it explicitly being not a SI unit.

The emissions used here, include the changes due to the implementation of IMO2020? It should be mentioned if it is expected an impact of this regulation on ozone formation due to shipping.

Page 4, line 18. Field rather than filed. In addition, why to use a one-year meteorology instead of the specific meteorology of each year? I believe that meteorological parameters have a strong influence on ozone formation and this is also what is mentioned in the conclusions.

Page 7, lines 25-26. This sentence seems to say that shipping is not relevant for ozone formation and it is opposite to what is said in conclusions.

Figure 1. What is the cause of the increment of emission in 2020? Fig. S2 does not show a significant increase of cargo throughput. Could it be simply related to the use of a different emission database?

Page 14, line 4 there is an “s” that should be eliminated.

This manuscript investigates the influence of shipping emissions on surface ozone concentrations in China. To support this analysis, the authors have extended the shipping emission inventory SEIMv2.0 for the year 2020. The study employs the WRF-CMAQ chemical transport model and the ISAM source apportionment module to assess the contributions of ocean-going, coastal, and river vessels to surface ozone concentrations. Additionally, a random forest machine learning model is applied to interpret the sensitivity of monthly mean ozone levels to various input features, including meteorological parameters, land-based anthropogenic emissions, and shipping-related emissions. While the study addresses a timely and important topic, several major concerns should be carefully addressed before the manuscript can be considered for publication.

A primary concern is the limited depth of analysis and clear contribution to current scientific knowledge on ozone pollution. The manuscript uses a style more aligned with a technical report, lacking a thorough link with the current literature on the role of shipping emissions in ozone formation, particularly in coastal and river basin environments. Furthermore, the novelty of the work and its broader implications are not clearly conveyed. A more robust discussion contrasting the study with recent literature studies conducted in other regions would significantly strengthen the manuscript’s relevance and potential impact.

Although the paper is generally well structured, several methodological aspects require further clarification. Notably, the stated objective of investigating interannual impacts of shipping emissions from 2016 to 2020 contrasts with the modeling setup, which uses meteorology from a single representative year (2018). This approach primarily assesses the impact of emission changes under fixed meteorological conditions, rather than capturing interannual variability. Emissions from other sources (e.g., international anthropogenic sources from 2010, and open burning from 2015) are also held constant. The implications of this modeling design should be explicitly acknowledged, and the study’s objectives reformulated to better reflect the actual scope of the simulations.

In this regards, the presentation of results as a five-year average raises concerns about the interpretation and relevance of the findings. It is unclear what this average represents given the modeling configuration. While averaging can simplify interpretation, it risks obscuring temporal variability and may lead to misleading conclusions about the spatial and seasonal influence of shipping emissions. I strongly recommend avoiding multi-year averaging in this context. Instead, results should be presented as sensitivity simulations under consistent meteorological conditions, with comparisons made between specific emission scenarios (e.g., 2016 vs. 2020). Furthermore, the analysis would benefit greatly from an angle considering high-ozone episodes (e.g., events when MDA8 > 120 µg/m³), as these events are of particular interest for air quality management. Moreover, with a more in-depth statistical analysis, the authors could examine the sensitivity and contributions of the different shipping sources during low-, medium-, and high-ozone concentration events.

The application of an explainable machine learning model to explore ozone drivers is an interesting addition; however, its implementation raises several issues. It is well established that meteorology significantly influences ozone formation, but hemispheric background ozone concentrations also play a crucial role, as highlighted in several recent studies (e.g., Jonson et al., 2018; Lupasçu and Butler, 2019; Shu et al., 2023; Garatachea et al., 2024). The omission of background ozone as a feature in the machine learning model is a significant limitation and likely biases the interpretation of feature importance. Given that the ISAM module is capable of capturing this background contribution, its integration into the machine learning framework should be considered. Additionally, the use of monthly mean concentrations limits the model’s utility for understanding episodic ozone dynamics, which often unfold over shorter timescales. A more granular temporal resolution would be more appropriate for exploring the drivers of ozone exceedances.

The manuscript would benefit from careful proofreading. While the general structure is acceptable, several sections require refinement for clarity and precision.

This manuscript addresses a topic of considerable scientific and policy interest. However, major revisions are required to improve the clarity of the methods, enhance the scientific discussion, and strengthen the novelty and relevance of the findings. I encourage the authors to address the comments above and the specific points provided below before considering the paper for publication.

Specific Comments:

• Page 1, Line 22: It is unclear what is considered an "effective ozone mitigation measure" in the context of China, and how controlling shipping emissions contributes, for example, to reducing ozone exceedances across the country. This should be clarified in the main text, preferably in the Introduction. A nationwide contribution of 3.5 ppb may be highly relevant if it leads to exceedances in specific regions.
• Page 2, Line 6: Please clarify whether ozone is emerging as a more significant issue in urban or rural regions across China.
• Page 2, Line 14: A reference should be provided for the reported reductions in ozone and its precursors in China. Is shipping now viewed as a major contributor because emissions from other sectors have already undergone significant reductions?
• Page 2, Line 21: Quantitative estimates of the increase in emissions should be provided.
• Page 2, Line 25: Quantitative results should be presented and compared with findings from similar studies conducted in other regions to support a more comprehensive literature discussion.
• Page 2, Line 40: Please include relevant references to support the statements made.
• Page 3, Line 6: Is the model resolution employed in this study sufficient to capture ozone exceedances across China? A brief discussion on the setup limitations and the rationale for its selection should be included in the Methods section.
• Page 3, Line 11: The use of machine learning to analyze the impact of shipping emissions on ozone formation should be better justified, especially considering this is the primary aim of the ISAM source apportionment tool. The added value of the machine learning approach relative to insights already provided by CMAQ-ISAM should be clearly explained.
• Page 3, Line 28: Please be more specific and avoid overly verbose statements.
• Page 3, Line 32: Clarify whether SEIMv2.0 is extended for 2020 only, and whether a recalculation for the 2016–2019 period was performed. The treatment of emissions beyond 200 nautical miles from the Chinese coastline should also be described. It is currently unclear which emission inventory and reference year are used for those sources (not shown in Table S2).
• Page 3, Line 36: Define the acronym "IMO" upon first use.
• Page 4, Line 4: Cite the regulatory measure that enforce the use of low-sulfur fuels in 2020. Additionally, could the observed sudden changes in OGV emissions in 2020 be partly attributed to the COVID-19 pandemic?
• Page 4, Line 12: The sentence should be rephrased to clarify that emissions from 2016 to 2020 are simulated using 2018 meteorology. The current wording is misleading. Also, clarify how annual and seasonal means are derived when only 4 out of 12 months are simulated. Justification for this simplification and its limitations should be provided.
• Table S1: Include a comprehensive caption that explains how the statistics are computed. Indicate which meteorological stations are used and describe the spatial and temporal aggregation methods.
• Page 5, Line 3: Please confirm whether chemical boundary conditions also correspond to the year 2018.
• Page 5, Line 5: Specify which tagging method within ISAM is used. The authors should justify the choice and discuss its implications, as different tagging schemes may yield significantly different results (see Shu et al., 2023).
• Page 5, Line 12: Clarify whether the 2018 evaluation is conducted using 2018 SEIM and MEIC emissions. The referencing of years throughout the manuscript is inconsistent and may cause confusion.
• Table S3: As with Table S1, improve the figure caption by clarifying how the metrics are calculated. Consider computing statistics based on MDA8 values rather than hourly data.
• Page 5, Lines 22–23: The Random Forest model is trained using CMAQ outputs, not ISAM. Including ISAM-derived information in the machine learning model could potentially enrich the analysis. Monthly averages may obscure key insights regarding the impact of shipping on peak ozone levels.
• Page 5, Line 28: This sentence highlights a key concern. If the main purpose of the Random Forest model is as stated here, the ISAM module already provides more direct and robust information. Please revise the sentence and clarify the rationale for applying machine learning.
• Page 5, Line 30: If results from 2016–2019 are used for training, does this imply that model predictions discussed in the Results section refer exclusively to 2020 emissions? Please clarify.
• Page 6, Line 14: Indicate whether the averages are calculated from hourly ozone values or based on MDA8. Since ozone concentrations at night are often overestimated in models, using hourly data may introduce bias. MDA8 is a more appropriate metric for evaluating daily peaks and understanding emission sensitivity. As noted earlier, presenting five-year averages under the current setup is potentially misleading.
• Page 6, Figure 3: Consider presenting total ozone concentrations and shipping contributions in separate panels using absolute values for clarity.
• Page 7, Line 25: Please explain why the relative contribution of shipping emissions appears higher in 2017 than in subsequent years, despite steadily increasing emissions (as shown in Figure 1). This suggests regional sensitivity that warrants further discussion.
• Page 8, Line 9: Include, in parentheses, the relative contribution of shipping to total ozone.
• Page 9, Line 17: To strengthen the analysis, consider presenting the full range (e.g., min, max, interquartile range) of shipping contributions, rather than just the mean. Contributions in specific regions may be substantial.
• Page 11, Line 3: No comments are made on seasonal variations in emissions. Do RV or CV emissions exhibit any significant seasonal patterns?
• Page 11, Line 5: The analysis presented may be incomplete due to the omission of hemispheric background ozone concentrations.
• Page 13, Line 8: This paragraph appears to question the robustness of the machine learning approach for analyzing ozone formation. Consider clarifying its intended role and limitations in this context.
• Page 13, Line 17: The conclusion section is currently too brief and does not convey the potential key findings of the study. Some conclusions (e.g., the role of temperature and solar radiation) are well known and may not constitute novel insights. The authors should more clearly explain the main findings and novelty of their work.

Technical Comments:

• The quality of several figures should be improved for readability and clarity.
• All figure and table captions should be self-contained and descriptive, clearly explaining the data presented.
• Page 1, Line 17: Replace “...mechanisms of shipping emissions...” with “...mechanisms by which shipping emissions...”.
• Page 2, Line 16: “volatile organic compounds”
• Page 2, Lines 22–23: “critically important”
• Page 2, Line 34: Ensure consistent terminology throughout the manuscript. Use either “ozone” or “O₃,” not both interchangeably.
• Page 2, Line 35: Replace “timeframes” with “periods.”
• Page 3, Line 3: The sentence is unclear; please revise for clarity and correct any typographical errors.
• Page 3, Line 8: Replace “allocate culpabilities of” with “apportion”
• Page 3, Line 18: Would not Wang et al. (2021) be the appropriate reference for SEIMv2.0?
• Page 3, Line 18: Remove the word “driven” after “by.”
• Page 3, Line 30: Use “VOC” instead of “HC.”
• Page 3, Line 38: Replace “IMO.” with “IMO;”
• Page 3, Line 40: Replace “RVs. (c) Finally, vessels” with “RVs; and (c) vessels.”
• Page 4, Line 15: Define the acronyms BRA, YRD, and PRD at first mention.
• Page 5, Line 8: Correct the citation typo.
• Page 6, Figure 2: Clarify what is plotted. Does each point represent the monthly average per grid cell?
• Page 7, Line 6: Correct the figure number “Figure SX.”
• Page 12, Line 11: The quality of the circular plot is too low to read the percentage values. The figure caption should explain the plot clearly, including the meaning of horizontal and vertical displacement in the cloud of points for each feature.
• Page 13, Line 23: “Although”
• Page 14, Line 4: Correct the typographical error.